CONFIDENCE.T Function (Gnumeric)

=CONFIDENCE.T(alpha,stddev,size)

Every argument is required; there are no documented optional arguments or default values. Gnumeric’s manual uses commas with a decimal point in these English examples; validate the installed interface’s locale settings before changing delimiters or numerical punctuation. The size argument does not accept the degrees of freedom directly: Gnumeric’s source computes its t critical value with n − 1 degrees of freedom internally. Statistics manual and source, checked 13 October 2026.

For data assumed to support a conventional confidence interval for a mean, with sample standard deviation s and at least 2 observations:

E = t_(1 - alpha/2, n - 1) * s / sqrt(n)

The two-sided confidence limits around sample mean x̄ are x̄ − E and x̄ + E. The t critical value is calculated with n − 1 degrees of freedom, not n. For the same numerical spread and sample size, the t margin will generally be wider than the corresponding normal margin because uncertainty from estimating σ must be accounted for. Gnumeric’s CONFIDENCE uses a normal quantile and population σ instead. Neither function judges whether the data are independent, representative, badly skewed or heavy-tailed. For a small sample, assumptions such as independent observations and an approximately normal underlying population deserve particular attention.

You summarized 36 independent measurements. The sample standard deviation is 12 and the sample mean is 100. Put these in Gnumeric:

The independent calculation uses t critical value t(0.975, 35) ≈ 2.0301079283, giving margin 4.0602158565 and interval approximately [95.939784, 104.060216]. For comparison, normal-based =CONFIDENCE(0.05,12,36) would have expected margin 3.9199279691; it is only methodologically appropriate if 12 represents the known population standard deviation, not merely a sample estimate. These numbers were independently calculated, not run in native Gnumeric.

Complete sample-data exercise (no precomputed mean or deviation)

Type the following values in A2:A13, one per row. They are small enough to reproduce without a file download:

You can also use =CONFIDENCE.T(0.05,STDEV(A2:A13),COUNT(A2:A13)) directly. Using STDEV is appropriate for the sample deviation; STDEVP applies a different denominator and must not be silently substituted. The paired interval endpoints illustrate arithmetic around the sample mean, not a range produced by one CONFIDENCE.T cell.

A 90% sensitivity check. =CONFIDENCE.T(0.1,8,25) is expected to return 2.7374113279, based on 24 degrees of freedom, whereas the normal-theory comparison with a truly known σ of 8 is 2.6317658031. For n=16 and s=5 at 99% confidence, =CONFIDENCE.T(0.01,5,16) is independently expected to return 3.6833911043. More demanding confidence levels or small n widen the half-width; this does not imply that larger margins prove the data are good.

Multiple independent groups. Prepare a separate t margin for each group’s count and sample deviation, rather than entering a grand total n for one group’s s. This function describes a confidence interval for a single population mean, not a significance test for the difference of two means or a paired-difference procedure. A two-group comparison needs a separately justified statistical method.

Automated range reporting. Keep the original observations, COUNT, AVERAGE, STDEV, confidence half-width and lower/upper endpoints in visibly labeled cells. Use the explicit formula for endpoints so readers can verify it and discover dropped blanks or nonnumeric values. If a survey or measurement process has dependencies, nonrandom selection, outliers or heavy skew, consider a statistical method appropriate for that data-generating process rather than simply trusting a spreadsheet output.

Too-wide interval? Check units, data entry errors, confidence level, sample size and whether STDEV is being applied to the correct observations. Different result from CONFIDENCE? The t and normal methods use different distributions and distinct meanings for stddev. Unexpected precision? Display formatting may round while keeping underlying numeric values. Blank/non-numeric cells? Verify the effective numeric sample count with COUNT, check missingness rather than treating blank as zero by assumption, and don’t mix a manually typed n with a filtered or edited sample. None of these troubleshooting hypotheses claims an application test was performed.

Gnumeric’s official public catalogue lists confidence.t as implementation Complete, but testing No Testsuite—distinct from Complete / Basic for confidence. The implementation entry means registered/source implementation, not a guarantee of native support in every build, packaged distribution, plugin configuration or platform. Release 1.12.62 is publicly announced, and the implementation details were independently checked in the checksum-verified official 1.12.62 release source. The Gnumeric Statistics manual labels this function Excel compatible; this run performed no workbook import/export, formula locale, Excel, LibreOffice, ODF/XLSX round-trip or native Gnumeric comparison tests. Treat other spreadsheet implementations as non-authoritative for Gnumeric and verify any migration in its intended installed version.

Read CONFIDENCE for known-population-σ normal margins; AVERAGE for estimating the center; STDEV for sample variability; STDEVP for a population deviation; and COUNT for numerical sample size.

• Gnumeric Statistics manual — CONFIDENCE.T: explicit argument order, sample deviation, truncated sample count, size and zero-deviation error cases, official Excel compatibility statement.
• Gnumeric public function catalogue: confidence.t implementation Complete, upstream testing No Testsuite; not proof of successful execution of the examples here.
• Gnumeric 1.12.62 release announcement: software release context, not an assertion of new/changed CONFIDENCE.T functionality in that release.
• Official Gnumeric 1.12.62 source archive: checksum-verified release plugins/fn-stat/functions.c was inspected for the required arguments, inverse-distribution formula and explicit sample-size/standard-deviation error branches. Source inspection is not native execution.
• Validation: Official source and documentation checked; SciPy quantiles and independent arithmetic performed. Native Gnumeric execution, screenshots, locale and workbook round-trip tests were not performed.